def test_fitness(self):
"""
Fitness must always be the same length as individuals
"""
pop = Population(individuals=[1, 2, 3, 4, 5])
# should raise error when trying to set fitness with length not
# equal to the length of indivuals
with self.assertRaises(PopulationFittnessDimensionError):
pop.fitness = [1, 2, 3]
# should allow setting fitness to an iterable with the same length
# as individuals
self.fitness = np.zeros(pop.npop)
self.assertEqual(len(self.fitness), pop.npop)
# setting fitness should force a sort of individuals and fitness
ind0 = np.random.rand(100, 5)
fit0 = np.arange(100)
pop = Population(individuals=ind0, fitness=fit0)
for v0, v1 in zip(ind0[0], pop.individuals[-1]):
self.assertEqual(v0, v1)
self.assertEqual(fit0.max(), pop.fitness[0])
# ... but should not change the shape of individuals
self.assertEqual(ind0.shape, pop.individuals.shape)
self.assertEqual(pop.fitness.shape, (100,))
# should not allow fitness to be greater than 1d
with self.assertRaises(PopulationFittnessDimensionError):
pop.fitness = np.zeros((pop.npop, 10))
def test_individuals(self):
"""
Updating individuals should force an update of fitness
"""
pop = Population(individuals=[1, 2, 3], fitness=[0, 0, 0])
self.assertEqual(pop.npop, 3)
# resetting individuals should force an update of fit
pop.individuals = [0, 1, 2]
with self.assertRaises(PopulationFittnessNotSetError):
f = pop.fitness
# npop should always be the length of individuals
ind0 = np.random.rand(100, 5)
pop = Population(individuals=ind0)
self.assertEqual(ind0.shape[0], pop.npop)
self.assertEqual(ind0.shape[0], len(pop.individuals))
self.assertEqual(len(ind0), len(pop.individuals))
def test_mutate(self):
"""
Should perform mutation on a subset of the population
"""
ind0 = np.asarray([1., 2., 3., 4.])
pop = Population(individuals=ind0, fitness=ind0)
# should change some values
pop.mutate(p=1.)
count = 0
for f0, f1 in zip(ind0, pop.individuals):
if f0 != f1:
count += 1
self.assertGreater(count, 0)
# should set fitness for changed individuals to nan
self.assertGreater(len(pop.inew), 0)
# should handle > 1d
ind0 = np.random.rand(50, 10)
pop = Population(individuals=ind0, fitness=range(50))
pop.mutate(p=1.)
self.assertGreater(len(pop.inew), 0)
# should not change shape of individuals
self.assertEqual(pop.individuals.shape, ind0.shape)
def test_crossover(self):
"""
Should perform binary crossover on a subset of the population
"""
ind0 = np.random.rand(4)
fit0 = np.arange(len(ind0))
pop = Population(individuals=ind0, fitness=fit0)
# should change some values
pop.crossover(p=1.)
count = 0
for f0, f1 in zip(ind0, pop.individuals):
if f0 != f1:
count += 1
self.assertGreater(count, 0)
# should set fitness for changed individuals to nan
self.assertGreater(len(pop.inew), 0)
# should handle > 1d
ind0 = np.random.rand(50, 10)
pop = Population(individuals=ind0, fitness=range(50))
pop.crossover(p=1.)
self.assertGreater(len(pop.inew), 0)
# should not change shape of individuals
self.assertEqual(pop.individuals.shape, ind0.shape)
# FIXME?? fails on np.log10(0) in power calc. in float2bin()
## should not produce NaNs
ind0 = np.round(np.random.rand(100))
pop = Population(individuals=ind0)
for p in [0.1, 0.5, 1.]:
pop.crossover(p=p)
self.assertEqual(len(np.nonzero(np.isnan(pop.individuals))[0]), 0)
def test_clone(self):
"""
Should create a new generation by copying the fittest indivuals in
the current generation
"""
fit0 = [1, 2, 3, 4]
pop = Population(individuals=fit0, fitness=fit0)
sumfit = []
for i in range(100):
pop.clone()
sumfit.append(np.sum(pop.fitness))
self.assertGreater(np.max(sumfit), np.sum(fit0))
# should not change shape of 1d array
ind0 = np.random.rand(50)
pop = Population(individuals=ind0, fitness=range(50))
pop.clone()
self.assertEqual(pop.individuals.shape, ind0.shape)
# should not change shape of multi-d array
ind0 = np.random.rand(50, 5)
pop = Population(individuals=ind0, fitness=range(50))
pop.clone()
self.assertEqual(pop.individuals.shape, ind0.shape)
# should not produce NaNs
ind0 = np.round(np.random.rand(100))
pop = Population(individuals=ind0)
for p in range(100):
pop.clone()
self.assertEqual(len(np.nonzero(np.isnan(pop.individuals))[0]), 0)
